Figure 1. College location for 2008-2016 data
Beyond the Playbook
Racial Diversity and Disparity in Football Quarterbacking
Abstract
In this study, we analyze racial disparities at different levels in competitive American football: High school, College, and the National Football League (NFL). In particular, we disentangle the selection process at different stages for the most pivotal position in a team: Quarterbacks.
Introduction
The intersection of race and the quarterback position in American football has been a topic of considerable discussion and analysis over the years [REFERENCES]. Historically, the quarterback position in the National Football League (NFL) was predominantly occupied by white athletes, reflecting broader societal attitudes and stereotypes about leadership and intellectual roles in sports. However, the landscape has been changing gradually.
Recent years have witnessed a significant shift in this trend. As of 2023, the NFL has seen a notable increase in the number of quarterbacks of African American and other racial backgrounds leading teams, challenging long-standing biases. For instance, at the start of the 2023 season, approximately 44% of starting quarterbacks in the NFL were African American, a record for the league and a marked increase from just 25% two decades earlier. This change not only reflects a shift in the racial dynamics within the sport but also indicates a broader societal progression towards inclusivity and recognition of diverse talents.
This study aims to delve deeper into the statistical trends of potential racial discrimination in the quarterback position, analyzing selection at different stages of a player’s career: from high school to college, and, potentially, from college to pro-ball.
Data
Sources
To analyze the progression of racial disparity in the quarterback position, we merge several data sources:
Quarterback Stats for College Football (1980 - 2016): This dataset contains scraped data from sports-reference.com about quarterbacks that play college football, including some key statistics such as number of games, passing yards, touchdowns, etc.
College Football Total QBR (2007-2019): Scraped dataset from espn.com, including 135 QBs per year, their ranking, and other statistics.
Player information for College Football (1980 - 2022): This is a dataset gathered by a Reddit user (see here), including information about their position, statistics, school and town of original, and some sparse NFL statistics as well.
Draft picks (1980-2022): Data set with all draft picks by year and position.
Maxpreps data for top high school quarterbacks (2012 - 2022): Information about top 30 high school quarterbacks each year (according to maxprep ranking), including school and statistics. This has been complemented by a manual search of the player’s current status and potential position change.
Perceived race classification (2007 - 2016): This is a variable manually gathered for QBs listed in the first source. This is a perceived race classification based on images of the players, with three distinct categories:
White,Black, andOther. Even though this variable does not capture the actual race of the player, it will be a proxy for race perception and how this aspect affects their selection at different stages.
We are currently working on harmonizing all these datasets, and adding high school demographic information for each player, in addition to extending the perceived race classification for additional years.
Data Description
College Level
In total, we have 509 individual college quarterbacks between the years of 2007 and 2016, that are distributed throughout the US as shown in the following map.
We also have the following characteristics for each of the players, as described in the following table:
Preliminary Results
College Level
First, we will compare the performance in College football between White quarterbacks and quarterbacks of color. From the following table, we can see that overall, non-White QBs perform similarly than White QBs. One key difference between both groups, though, is that non-White QBs have significantly higher rush attempts and rush yards in comparison with passing attempts and passing yards, which could also demonstrate a difference in playing style between both. In terms of touch downs, however, both groups have similar performance.
| Mean | Std. Dev. | Mean | Std. Dev. | Diff. in Means | p | |
|---|---|---|---|---|---|---|
| Year | 2013.48 | 2.05 | 2013.43 | 2.11 | -0.05 | 0.82 |
| Rank | 64.50 | 35.52 | 61.05 | 38.56 | -3.45 | 0.37 |
| QBR | 56.44 | 16.37 | 58.96 | 17.81 | 2.52 | 0.17 |
| Raw Rank (ESPN) | 55.07 | 15.06 | 56.76 | 16.16 | 1.69 | 0.31 |
| Games | 11.24 | 2.03 | 11.66 | 1.85 | 0.43 | 0.03 |
| Plays | 452.19 | 119.03 | 456.21 | 120.31 | 4.02 | 0.75 |
| Completions | 205.10 | 72.07 | 189.11 | 66.16 | -15.98 | 0.02 |
| Points contributed | 13.41 | 30.60 | 17.18 | 34.23 | 3.76 | 0.28 |
| Passing TDs | 17.66 | 9.18 | 17.41 | 8.77 | -0.25 | 0.78 |
| Passing TDs per Game | 1.53 | 0.66 | 1.46 | 0.64 | -0.07 | 0.28 |
| Pass Attempts | 338.57 | 103.30 | 315.80 | 91.51 | -22.77 | 0.02 |
| Pass Yards | 2493.96 | 929.13 | 2357.90 | 862.21 | -136.05 | 0.13 |
| Yards Per Attempt | 7.27 | 1.06 | 7.36 | 1.20 | 0.10 | 0.41 |
| Avg Yards Per Attempt | 7.06 | 1.54 | 7.14 | 1.74 | 0.08 | 0.63 |
| Pass | 35.31 | 24.76 | 31.96 | 24.99 | -3.35 | 0.20 |
| Rush Attempts | 68.79 | 40.47 | 108.02 | 48.75 | 39.23 | <0.01 |
| Rush Yards | 126.39 | 258.08 | 388.01 | 328.19 | 261.62 | <0.01 |
| Avg Rush Yards | 0.77 | 2.79 | 3.09 | 1.88 | 2.32 | <0.01 |
| Run | 4.86 | 9.71 | 11.54 | 13.14 | 6.68 | <0.01 |
| Sack | -12.54 | 5.49 | -12.60 | 4.92 | -0.06 | 0.91 |
| Pen | 1.88 | 2.28 | 1.66 | 2.34 | -0.22 | 0.36 |
| Interceptions | 8.92 | 3.44 | 8.75 | 3.47 | -0.17 | 0.63 |
| Drafted | 0.22 | 0.41 | 0.16 | 0.37 | -0.06 | 0.14 |
| Pick | 119.10 | 81.85 | 90.05 | 75.00 | -29.05 | 0.13 |
| Round | 3.98 | 2.15 | 3.10 | 1.97 | -0.88 | 0.08 |
Figure 2. SD Difference in QB charactersitics by race (Non-White vs White)
Looking at the distribution of performance, we can observe that QBs of color follow a bi-modal distribution, but with an average performance in terms of QB rating that is similar (or even slightly higher) than White quarterbacks.
We observe a similar trend when analyzing points contributed as a function of QB rating. While there is a clear positive association between both, there does not appear to be a significant difference in terms of distribution by race.
In terms of their professional prospects, though, measured as being drafted by an NFL team as a quarterback, results differ between both groups. The following models shows the difference in probability of being drafted (NonWhiteQB) for Non-White QBs compared to White QBs. Even after controlling for year and some performance metrics, such as QB rating and passing touch downs, the differences persist:
| Drafted | Drafted | Drafted | Drafted | |
|---|---|---|---|---|
| (Intercept) | −13.727 | 0.376 | −5.146 | −2.715 |
| (18.075) | (17.502) | (17.304) | (16.949) | |
| NonWhiteQB | −0.056 | −0.088* | −0.096** | −0.070* |
| (0.039) | (0.036) | (0.036) | (0.035) | |
| Year | 0.007 | 0.000 | 0.002 | 0.001 |
| (0.009) | (0.009) | (0.009) | (0.008) | |
| QBR | 0.009*** | 0.007*** | 0.004** | |
| (0.001) | (0.001) | (0.001) | ||
| Games | 0.028** | −0.002 | ||
| (0.010) | (0.010) | |||
| Passing_TDs | 0.015*** | |||
| (0.003) | ||||
| Num.Obs. | 509 | 484 | 484 | 484 |
| R2 | 0.005 | 0.130 | 0.144 | 0.201 |
| R2 Adj. | 0.001 | 0.125 | 0.137 | 0.193 |
| + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001 |
Given that, on average, a College QB has a 20% probability of being drafted in our dataset, a negative effect between 7 and 9.6 percentage points on being drafted represents a decrease in probability between 35% and 48% associated to being a QB of color.
Given that White and Non-White QBs could also have different styles of play, where one could be better rewarded because of their play style, we re-weight non-White quarterback observations to match the style of play of White QBs in terms of passing yards/attempts, rushing yards/attempts, interceptions, touch downs, games played, and other aggregated performance characteristics such as QB Rating, raw rating, and points contributed.
We follow Chattopadhyay, Hase, and Zubizarreta (2020) for building stable weights that approximately balance the empirical distribution of the aforementioned covariates with a tolerance of 0.01 SD, using the distribution of performance for White QBs as the target. The following table shows the balance between both groups after weighting.
| White QB | Non-White QB | Target | Diff in Means (SD) | |
|---|---|---|---|---|
| Games | 11.38 | 11.40 | 11.38 | 0.01 |
| Completions | 209.08 | 209.74 | 209.08 | 0.01 |
| Pass Attempts | 344.58 | 343.67 | 344.58 | 0.01 |
| Pass Yards | 2546.85 | 2540.87 | 2546.85 | 0.01 |
| Avg Yards Per Attempt | 7.11 | 7.13 | 7.11 | 0.01 |
| Interceptions | 9.02 | 9.05 | 9.02 | 0.01 |
| Rush Attempts | 70.39 | 70.86 | 70.39 | 0.01 |
| Avg Rush Yards | 0.88 | 0.90 | 0.88 | 0.01 |
| QBR | 56.44 | 56.62 | 56.44 | 0.01 |
| Pass | 35.31 | 35.06 | 35.31 | 0.01 |
| Plays | 452.19 | 453.39 | 452.19 | 0.01 |
| Points contributed | 13.41 | 13.76 | 13.41 | 0.01 |
| Raw Rank (ESPN) | 55.07 | 54.91 | 55.07 | 0.01 |
Finally, comparing the probability of being drafted for both groups after weighting, we find that there is a 0.077 (SE = 0.032; p-value = 0.0769) decrease in the probability of being drafted for QBs of color compared to White QBs, which is similar to our previous results. These findings indicate that even though there could be other potential confounders in play, playing style as captured by these main metrics are not the ones driving this significant difference in drafting probability.